The protocols were approved by the University of Pennsylvania Institutional Animal Care and Use Committee.

All horses underwent routine dental and foot care and scheduled administration of appropriate vaccinations and de-worming agents. The horses were no longer actively racing. These horses were pasture fit and not exercised on a routine basis. Based on physical examination and routine blood chemistry analysis, all the horses in the study were in good health. Horses were housed on pasture and brought into temperature controlled stalls 7 days before ESWT. Twelve horses were administered a total of 2700 impulses at an energy density of 0.55 mJ/mm² to one of the dorsal metacarpal bones using a ESWT unlimited, Duolith SD1 T-Top, Vet, F-SW (STORZ MEDICAL AG, Tägerwilen, Switzerland). Blood samples (10 mL aliquots) were collected in heparinized tubes by venipuncture of the jugular vein before ESWT (0 h) and at 24, 48 and 72 h post ESWT.

RNA was extracted from fresh blood samples using LeuKoLOCK™ total RNA isolation procedure (Applied Biosystems, Foster City, CA, USA) as previously described [18] . Genomic DNA was eliminated by RNase-free DNase I digestion during the isolation procedure. The quality and quantity of extracted RNA samples were assayed using the Agilent Bioanalyzer system (Agilent Technologies, Inc., Santa Clara, CA, USA). All samples were quantified using a NanoDrop 2000 spectrometer (Thermo Fisher Scientic, Pittsburgh, USA). Complementary DNA (cDNA) was synthesized using hexanucleotide primers and Multiscribe Reverse Transcriptase (Applied Biosystems, Foster City, CA, USA) in a final reaction volume of 20 µL containing 2 µL reverse transcriptase buffer, 5.5 mM MgCl₂, 0.5 mM deoxyribonucleoside triphosphate, 2.5 mM random hexamers, 0.4 U RNase inhibitor, 1.25 U Multiscribe Reverse Transcriptase and 200 ng RNA. All reactions were performed under the following conditions: initial primer incubation step was performed for 10 min at 25˚C, followed by reverse transcriptase incubation for 2 h at 37˚C, terminated by reverse transcriptase inactivation for 5 min at 95˚C, and cooled to 4˚C before storage at −80˚C pending analysis.

Twelve RGs were selected from commonly used in equine research [8] -[10] and commercially available products (Table 1). All primers were designed with Primer Express 3.0.1 (Applied Biosystems, Foster City, CA, USA) to amplify at 60˚C and to bind specifically to different exons of equine cDNA sequences. The qPCR efficiency ranged from 92% - 99% (Table 2).

Amplification of specific PCR products was detected using SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, USA). All primers employed were custom-synthesized by Applied Biosystems, Inc. Amplification efficiency of each primer pair was evaluated by the standard curve method. Gene expression study for all twelve genes was performed in 96-well plates using ABI 7500 real time PCR System (Applied Biosystems, Foster City, CA, USA). PCR step was carried out in 20 μL, including 10 μL 2x SYBR Green PCR Master Mix, 50 nM of each primer and 5 μL cDNA samples. The reaction mixtures were heated at 95˚C for 10 min and then subjected to 40 cycles at 95˚C for 15 sec, and 60˚C for 1 min. Each PCR reaction was performed in triplicate paralleling with no template control.

qPCR data were analyzed using SDS Software, and standard curves were fit using linear regression (v 2.0.6; Applied Biosystems, Foster City, CA, USA). Real-time PCR amplification efficiencies were calculated using the

slope of the standard curve.

All analyses were performed using SAS 9.3 (SAS Institute Inc. Cary, NC, USA). The normality test was performed by the Shapiro-Wilk method combining a graphic method: normal probability plot. Data were presented as mean ± SD. Pair-wise comparisons of gene expression levels among four groups of subjects were performed using one-way ANOVA when expression levels were normally distributed, and a non-parametric (Wilcoxon) test was applied when expression levels were not normally distributed. Bonferroni correction was used to adjust for multiple comparisons. Results were considered statistically significant when p < 0.05.

The stability of the selected RGs and the number of optimal RGs were determined using SASqPCR [19] , a SAS macro-program that was developed according to the algorithm of geNorm [13] .

For comparison of RGs transcription levels, the quantification cycle (C_q) values were plotted directly, assuming the same threshold for all genes evaluated. C_q is defined as the number of cycles needed for fluorescence to reach a specific threshold of detection and is inversely correlated with the number of RNA template present in the reaction. The twelve RGs analyzed in this study displayed a wide range of expression, with C_q values between 17 and 26 (Figure 1). The normality of C_q values for all the genes evaluated was tested using the ShapiroWilk method combining a graphic method: normal probability plot. Five genes, ACTB, HSP90A, HPRT1, TBP and UBC showed normal distribution in all samples collected at different time points. The remaining seven genes did not show normal distribution in one or more samples.

The twelve RGs were clearly distributed into different categories of expression level. Most genes coded for highly expressed mRNAs with the majority of C_q values between 17 and 22 cycles; five other genes, GUSB, SDHA, H6PD and HPRT1 and TBP, were moderately expressed with C_q values between 23 and 26 cycles. It should be noted that TBP had slightly higher expression levels in the 72 h samples than in the remaining samples (Figure 1).

As suggested by Ohl et al. [16] , pairwise comparisons on raw C_q values among 4 different samples were performed, and there were significant differences in gene expression of TBP between 72 h and the other 3 samples (72 h vs. control. p = 0.0003; 72 vs. 24 h, p = 0.0002; 72 vs. 48 h, p = 0.0004). These results indicated variability in TBP gene expression level by ESWT, even though it had previously been considered a house-keeping gene.

The main objective of this study was to identify RGs in the white blood cells with minimal variability following treatment of horses with ESWT. In order to determine the least variable RGs, we evaluated expression stability of the twelve candidate RGs in the panel of samples using SASqPCR. Gene expression stability measurement (M) was calculated based on the average pairwise variation between a particular gene and all other genes studied. The suggested cut-off value of M was 1.5 [13] .

In the control samples, M values for all genes were between 0.2 and 0.6, and the expression stability from the most stable to the least stable was as follows: HPRT1 > ACTB > HSP90A > SDHA > GUSB > B2M > UBC > NONO > TBP > H6PD > RPL32 > GAPDH (Figure 2(a)).

In the 24 h samples, the M values for all genes were also between 0.2 and 0.6, and the order of expression stability from the most stable to the least stable was as follows: ACTB > HSP90A > HPRT1 > TBP > GUSB > H6PD > B2M > NONO > GAPDH > SDHA > UBC > RPL32 (Figure 2(b)).

In the 48 h samples, the M values for all the genes were still between 0.2 and 0.6, the expression stability from the most stable to the least stable was as follows: ACTB > HPRT1 > HSP90A > GUSB > H6PD > SDHA > RPL32 > TBP > NONO > UBC > GAPDH > B2M (Figure 2(c)).

In the 72 h samples, M values for all genes were still between 0.2 and 1.6, showing a much wider range compared to the rest of the samples, the expression stability calculated for the genes analyzed was from the most stable to the least stable: ACTB > HPRT1 > H6PD > NONO > RPL32 > SDHA > GAPDH > HSP90A > GUSB > UBC > B2M > TBP (Figure 2(d)). It should be noted that the M value for TBP was 1.505, which was above the cut-off value of 1.5, suggesting that TBP was not a suitable reference gene in the 72 h samples. This was consistent with the pairwise comparison results.

To determine the number of stable RGs needed for normalization, a pairwise variation V_n/n+1 (n = number of RGs) was calculated for two sequential normalization factors (NF) NF_n and NF_n+1 for all samples. A large V value indicates that the added gene had a significant effect and should probably be included for calculation of the normalization factor. In the present study, the relative quantities of all candidate RGs in all samples were used to calculate V. As shown in Figure 3, all the pairwise variations, V_n/n+1, were below the 0.15 cut-off value [13] , an additional RG did not significantly contribute to improving accuracy. Since V_1/2 values for all samples were less than 0.15, one most stable RG was sufficient for accurate calibration, and an additional RG was not required as an internal control.